Artificial intelligence has captured the world’s imagination with impressive advances, from the emergence of generative models like ChatGPT to the promise of automating complex tasks in key industries.
Recently, a new actor has emerged with force: AI agents. Companies like HubSpot, Salesforce and Microsoft have presented their new solutions based on these intelligent assistants, promising to revolutionize the management of tasks and workflows.
But are we really facing a disruptive technology or simply experiencing a new cycle of technological hype?
What are AI agents?
AI Agents are systems designed to carry out tasks autonomously, without the need to constantly receive instructions from users.
Unlike generative AI models, which require continuous prompts to generate responses or content, AI agents have the ability to execute actions automatically, managing several stages of a process simultaneously or sequentially.
In essence, they can be seen as “employees” who not only perform repetitive tasks, but can also learn and adapt to circumstances.
For example, an AI agent at a telecommunications company could automatically handle a customer support request.
From categorizing the query, to verifying customer data, to resolving the problem or assigning a technician, the agent could carry out all these actions without human intervention.
The promise of AI agents
The appearance of these agents has been accompanied by great expectations.
During HubSpot’s Inbound event, its CTO, Dharmesh Shah, imagined a future where there would be an AI agent for every possible task in marketing, sales and customer service, suggesting it could be as disruptive as the arrival of mobile apps.
Indeed, Shah’s vision resonates with the idea that AI agents will not only execute existing tasks, but also create new agents for tasks we cannot yet imagine.
The business vision is clear: automate workflows, improve productivity and reduce the need for human intervention in everyday processes.
Instead of having employees dedicated to answering simple queries, managing databases, or making initial sales, AI agents would take on much of this work, allowing employees to focus on more strategic tasks.
Hype or reality?
Despite the promises, not everyone is convinced that AI agents will live up to these expectations.
Some critical voices, such as those expressed by Tegan Jones in her recent article “Is the AI Agent Hype Just Gen AI All Over Again?”, point out that the overwhelming enthusiasm for generative AI has given way to a less spectacular reality.
While generative AI has certainly transformed aspects of work and communication, reality has shown that it is not the magic bullet that some had anticipated. Jones suggests that something similar could be happening with AI agents.
Companies like Microsoft, Salesforce, and HubSpot are bringing these agents to market with ambitious promises, but their actual short-term capabilities may be overstated.
While they can be useful for automating routine tasks and improving productivity in defined areas, it is still unclear whether they will be able to meet expectations in the long term, especially when it comes to managing more complex tasks that require human judgment.
The technical and ethical challenges of AI agents
Like any emerging technology, AI agents are not without challenges.
In his most recent analysis, IBM’s Cole Stryker highlights four “crises” that could limit the development and adoption of this technology: the data crisis, the computing crisis, the energy crisis, and the use case crisis.
These factors represent serious obstacles to the expansion and effectiveness of AI agents.
- Data crisis: AI agents require enormous amounts of data to train. However, data sources are increasingly restricted for privacy and intellectual property reasons, which could limit the development of agents that can work in diverse scenarios.
- Computational crisis: Processing these agents requires significant computational resources, which is leading to a shortage of specialized chips, such as GPUs, which are necessary to train advanced models.
- Energy crisis: Current AI models consume large amounts of energy, and the global energy infrastructure may not be prepared to support mass adoption of AI agents without affecting the environment.
- Use case crisis: Although AI agents promise to transform the company, applications that fully justify the multi-million dollar investments that companies are making in their development have not yet been identified.
These obstacles are not insurmountable, but they require a pragmatic and careful approach in their implementation.
What is the current reality of AI agents?
Despite the challenges, AI agents are already proving useful in certain contexts, such as customer service, marketing, and sales.
As Chris Hay, distinguished engineer at IBM, points out, we are in a correction phase after the initial rush for generative AI. Companies should have realistic expectations about what AI agents can achieve right now.
We are likely to see AI agents operate effectively in well-defined and controlled areas, where workflows are repetitive and can be automated without constant human intervention.
However, in more complex tasks that require human judgment, technology is not yet up to the task of replacing people.
Additionally, AI agents will not operate in isolation. As Brent Smolinksi, also from IBM, explains, AI agents must be part of a broader technology strategy, combining multiple types of AI models and other technologies to effectively solve business problems.
It is not just a matter of using large language models, but of taking advantage of a more diverse set of tools adapted to each specific situation.
Conclusion: Between hype and reality
While these agents have the potential to automate repetitive tasks and increase efficiency, it’s important not to fall into the same hype cycle we’ve seen with generative AI.
Companies must take a balanced and pragmatic approach: identify areas where AI agents can have a real impact on their productivity and apply the technology strategically, without expecting silver bullets.
Instead of succumbing to the hype, it is crucial to critically evaluate the benefits and limitations of this technology based on the specific needs of each organization.
This post is also available in: